A Multi-component CNN-RNN Approach for Dimensional Emotion Recognition in-the-wild

  • 2018-05-03 17:54:44
  • Dimitrios Kollias, Stefanos Zafeiriou
  • 42

Abstract

This paper presents our approach to the One-Minute Gradual-EmotionRecognition (OMG-Emotion) Challenge, focusing on dimensional emotionrecognition through visual analysis of the provided emotion videos. Theapproach is based on a Convolutional and Recurrent (CNN-RNN) deep neuralarchitecture we have developed for the relevant large AffWild Emotion Database.We extended and adapted this architecture, by letting a combination of multiplefeatures generated in the CNN component be explored by RNN subnets. Our targethas been to obtain best performance on the OMG-Emotion visual validation dataset, while learning the respective visual training data set. Extendedexperimentation has led to best architectures for the estimation of the valuesof the valence and arousal emotion dimensions over these data sets.

 

Introduction (beta)

None

 

Conclusion (beta)

None